Why did Datadog growth slow in 2024-25?

The renewal meeting where the deceleration becomes visible

Growth decelerations at consumption-priced companies rarely announce themselves in a churn report. They show up in a room that looks, on the surface, like a win. Picture a Datadog account manager walking into a Q3 FY24 renewal with a customer spending roughly $1.4M a year across infrastructure monitoring, APM, and logs. Nobody in the room is unhappy. The engineering org loves the product. The platform team has 900 dashboards and every on-call runbook links back to a Datadog notebook. There is no competitive displacement risk that anyone can name out loud.

And yet the customer signs for $1.35M. Not because they found a cheaper vendor, but because their FinOps lead spent the previous quarter reading ingest reports and found that 40% of their log volume came from three debug-level services nobody has queried in eleven months. They dropped log retention on those from 30 days to 7. They cut about 2,000 custom metrics whose cardinality had exploded through a Kubernetes label they never intended to index. They consolidated four staging environments into one. None of that is churn. None of it is a competitive loss. It is a customer getting better at buying — and on a consumption model, a customer getting better at buying is a revenue event.

Multiply that room by a few thousand accounts and you get the FY24-FY25 shape. Revenue grew from roughly $2.1B in FY23 to about $2.7B in FY24 to about $3.1B in FY25 — real, compounding, enviable growth by almost any standard. But the *rate* stepped down every few quarters: ~27% to ~26% to ~25% to ~24% to ~23% by Q4 FY25, with the FY26 guide implied in the low-20s, the first sub-25% guide the company had issued. The absolute dollar adds stayed enormous. The percentage stopped cooperating.

This is the frame that matters for anyone doing RevOps work on a consumption business: the deceleration was not a demand failure. Logo counts kept climbing. The AI-native cohort was called out repeatedly as the fastest-growing customer segment. What changed was the *intensity per existing dollar* — how much more each retained customer chose to consume year over year. That variable is net revenue retention, and it is the load-bearing number in the entire story. When management said on the Q3 FY24 call that "a couple of large customers" optimized usage more aggressively than expected, the stock fell roughly 10% intraday not because two accounts matter to a $2.7B revenue base, but because investors correctly read it as a signal about the behavior of *every* account.

The adjacent lesson generalizes past Datadog. Any vendor whose meter runs on volume — Snowflake on compute credits, Twilio on messages, MongoDB Atlas on cluster hours, most modern data infrastructure — has the same structural exposure. The product is priced to grow when the customer grows, which is beautiful in an expansion cycle and unforgiving in an efficiency cycle. The 2023-2025 window was an efficiency cycle. AWS growth fell from north of 30% into the low teens over roughly the same period; that was not a coincidence happening beside Datadog's slowdown, it was the same wave hitting a different point on the same stack.

How consumption revenue actually decelerates

The mechanism is worth unpacking carefully, because "customers spent less" hides how the math actually propagates through a P&L and a forecast.

Start with the unit. Datadog charges per host for infrastructure, per ingested and indexed gigabyte for logs, per custom metric, per APM host and span volume, per synthetic test run, per session for RUM. Each of those meters is downstream of an engineering decision that no salesperson touches. When a platform team moves from 40 always-on EC2 instances to a Kubernetes cluster with aggressive horizontal pod autoscaling, host count can drop even as the workload grows. When a team switches a chatty service from DEBUG to WARN, indexed log volume falls by an order of magnitude overnight. When someone finally deletes the pod_name label from a custom metric, the cardinality collapse can remove tens of thousands of billable time series in a single deploy.

Now layer in the second-order effect. Consumption revenue is recognized as it's used, so an optimization taken in month two of a contract shows up immediately in reported revenue and then *repeats* for every remaining month. It's not a one-quarter dent; it resets the baseline the next renewal negotiates from. That's why a wave of optimizations across a customer base produces a step-down in growth rate rather than a single bad quarter — the effect is cumulative and it compounds against the following year's comparison.

Third, watch how it interacts with the customer mix. Datadog's disclosed cohort of customers paying $100K+ in ARR crossed roughly 3,500. That group carries the large majority of revenue. Optimization discipline is not evenly distributed — it correlates almost perfectly with spend. The $30K customer has no FinOps function and will never build one. The $3M customer has a named cloud cost owner, a quarterly review, and a chargeback model that makes every engineering manager care about their own observability line item. So the optimization wave hit precisely the accounts that carry the revenue, which is why a modest-sounding behavior change produced a visible growth-rate change.

The feedback edge from J back to G is the part most models miss. A reset baseline doesn't just lower this year's number, it makes next year's expansion arithmetic harder, because the denominator you're expanding from was already trimmed of its slack. That is the structural difference between a demand shock and an efficiency shock: demand comes back, efficiency doesn't unwind. Nobody re-adds 2,000 useless custom metrics once they've been deleted.

The fourth mechanism is pricing structure. As the optimization wave hit, Datadog leaned harder on annual and multi-year commit pricing — customers trading a volume commitment for a meaningful per-unit discount. Commit pricing is strategically correct: it converts volatile consumption into forecastable revenue and materially improves renewal visibility. But it has a near-term cost. A customer who was paying rack rate on overage now pays a discounted rate on committed volume, so realized revenue per unit falls even when unit volume is flat. In RevOps terms, you improve the quality of revenue while temporarily suppressing its reported growth rate. Both things are true at once, and the market prices the second one.

The numbers that matter, and the ranges around them

Specific figures anchor the argument, so here are the ones with real disclosure or well-established directional support behind them, plus honest labeling of what's an estimate.

Revenue and rate. FY23 finished around $2.1B, FY24 around $2.7B, FY25 around $3.1B. Quarterly growth walked down roughly: ~27% in Q1 and Q2 FY24, ~26% in Q3 FY24, ~25% in Q4 FY24 and Q1 FY25, ~24% in Q2 FY25, ~23% in Q3 and Q4 FY25. The FY26 guide implied low-20s. Anyone modeling this should note the pattern: one point of deceleration roughly every two quarters, remarkably linear, which is itself a signal that the driver was structural rather than event-driven. Event-driven slowdowns are lumpy. Structural ones are smooth.

Net revenue retention. This is the single most informative line. FY22 sat above 130%. FY23 came down near 120%. FY24 and FY25 ran around 115%. Do the sensitivity: on a stable base with no new logos at all, growth is approximately NRR minus 100%. Going from 130% to 115% mechanically removes about 15 points of growth from the retained base. That Datadog only decelerated by three or four points at the total-company level means new logo acquisition and new SKUs absorbed most of the damage. That's the underappreciated part of the story — the top of the funnel was doing heroic work to mask what was happening in the installed base.

Customer cohorts. Roughly 3,500 customers above $100K ARR. Growth in that cohort slowed from the mid-20s percent annually to something closer to the mid-teens. The law of large numbers is doing exactly what it does: to add 25% to a 3,500-logo base you need ~875 net new six-figure customers in a year, versus ~250 when the base was 1,000. The market for organizations that will spend $100K+ on observability is large but not infinite, and the most obvious ones were landed first.

Segment shape. US revenue runs roughly 70-75% of the total, with international in the mid-20s to low-40s depending on how you cut it across reporting periods. EMEA growth decelerated meaningfully as the region's cloud adoption curve matured. A stronger dollar in 2024 added a couple of points of translation headwind on non-USD contracts — small, but it lands right at the margin where a 24% print becomes a 23% print.

The offsetting lines. Cloud SIEM crossed a $100M ARR run rate and was growing north of 50%, off a small base. Bits AI, announced in late 2024 as the AI-workload observability layer, was not materially revenue-generating in FY25 — call it under 5% of revenue on any reasonable estimate. Both matter enormously to the FY27-FY30 narrative and almost not at all to the FY25 number. That gap between narrative importance and current-period materiality is exactly where investor disagreement lives.

Sales efficiency. Directionally, the self-serve motion that drove a large share of early logo acquisition contributed proportionally less over time as the cloud-native startup market saturated, and enterprise sales headcount grew to compensate. Enterprise motion means longer cycles — think 75-90 days where a PLG-sourced deal closed in 45-60 — and higher acquisition cost per dollar of ARR. Treat the specific ratios as estimates rather than disclosure, but the direction is well-supported and it's the standard maturation pattern: PLG gets you to a few hundred million, enterprise sales gets you past a billion, and the transition always shows up as temporarily worse efficiency.

The comparison set. A 24% growth rate on a $3B base is not an outlier among enterprise software companies at scale. Salesforce, Workday, and ServiceNow all decelerated meaningfully as they crossed the $3B mark. The reason Datadog's deceleration felt sharper is that consumption pricing had made its *acceleration* look sharper on the way up. Volume-metered models overshoot in both directions relative to seat-based ones — they capture more upside when customers expand and give more back when customers optimize. If you're a RevOps leader choosing a pricing model, that volatility is the real trade, and it's worth deciding deliberately rather than discovering it in a down cycle.

What Datadog could trade away, and what each trade costs

Every available response to a consumption slowdown has a bill attached. Laying them side by side is more useful than picking a favorite.

Push harder on commit contracts. You gain forecastability, renewal visibility, and a defensible floor under revenue. You pay in discount — 15-25% off list is a common shape for multi-year commitments — and you pay in reported growth, because committed volume is priced below the overage rate it replaced. You also introduce a new risk: an over-committed customer whose usage falls short arrives at renewal angry and negotiating from strength. Commits that customers can't consume become renewal-cycle liabilities.

Move upmarket aggressively. Larger deals, bigger logos, more revenue per rep. The cost is that the largest enterprises have the most sophisticated procurement, run competitive bake-offs as standard practice, and have real leverage. They also have the strongest FinOps functions — meaning the accounts you most want are the accounts most likely to optimize you. Moving upmarket to escape a consumption slowdown is partly running toward the source of it.

Bundle and cross-sell into adjacent SKUs. This is the Cloud SIEM and Bits AI path, and it's the most durable of the options because it creates new meters rather than re-pricing existing ones. The cost is time and focus. A security SKU means competing with CrowdStrike and Palo Alto, which is a different buyer, a different sales motion, and a different compliance surface. New SKUs also take years to become material — Cloud SIEM crossing $100M ARR is genuinely impressive and still roughly 3% of revenue.

Defend on price against the hyperscalers. AWS CloudWatch, Azure Monitor, and GCP Cloud Operations are bundled into commitments enterprises are already making, which makes them structurally cheap. Grafana's open-source-anchored model wins on cost in a meaningful share of mid-market evaluations. Datadog's counter has always been that the integrated platform is worth the premium — one agent, one query language, one correlation layer across metrics, traces, and logs. That argument holds well in complex environments and holds poorly for a team that only needs metrics on twelve services. Competing on price against a bundled hyperscaler is a fight you can't win on the merits of the invoice; you win it, if at all, on total cost including the engineering hours a stitched-together stack consumes.

Do nothing structural and ride it out. Underrated. Optimization waves exhaust themselves — there's a finite amount of waste in any telemetry pipeline, and once it's cut, growth reverts to underlying workload growth. The cost is a few years of compressed multiple and the risk that competitors use the window to establish footholds in accounts.

The realistic answer, and the one the disclosed behavior points to, is a mix weighted toward commits and adjacent SKUs. That's a strategy that trades near-term reported growth for durability, which is usually the right call for a company with a decade-long horizon and the wrong call for a stock priced on next year's number. Both audiences are being served by the same set of decisions, and they will grade it differently.

Pitfalls when you're reading — or living through — a slowdown like this

If you run revenue operations at a consumption-priced company, or you're evaluating one, these are the errors that cost the most.

Reading a growth-rate decline as a demand problem. The first instinct when growth slows is to inspect pipeline, and pipeline is usually fine. Separate your growth into three buckets before diagnosing anything: new logo revenue, expansion within retained accounts, and contraction plus churn. Datadog's story lives almost entirely in bucket three bleeding into bucket two. If you fire your CMO over a consumption-optimization slowdown, you've solved the wrong problem and lost a year.

Treating NRR as one number. Blended NRR hides everything interesting. Break it out by cohort — segment, tenure, product mix, region. A blended 115% could be a $1M+ cohort at 105% and an SMB cohort at 140%, which implies a completely different playbook than the reverse. Also separate gross retention from net: if gross retention is holding at 95%+ and net has fallen, you have a consumption problem, not a satisfaction problem, and consumption problems are addressed with packaging and commits rather than customer success headcount.

Missing the optimization wave because you only watch churn. Logo churn dashboards will show green through an entire deceleration. Instrument leading usage signals instead: week-over-week ingest volume by account, custom metric counts, retention-window configuration changes, host count trend. A customer who cuts log retention this month is telling you what their renewal looks like nine months out. Most vendors have this telemetry about their own customers and never route it to the revenue team.

Selling commits without consumption modeling. The fastest way to turn a good commit strategy into a bad one is to size the commitment off the customer's optimism instead of their trend line. Model committed volume against actual trailing usage with an explicit growth assumption you write down, then size the commit to something the customer will comfortably exceed. An overage conversation is a pleasant one. A shortfall conversation at renewal is where discounts go to die.

Assuming new SKUs rescue the current fiscal year. Cloud SIEM at $100M ARR growing 50% adds roughly $50M — real money and roughly two points of growth on a $2.7B base, at best. New products fix the three-year story, not the four-quarter one. Any model that leans on a new SKU to hold a near-term growth rate is a model that will miss.

Ignoring the pricing-model choice upstream. The broader RevOps lesson: consumption pricing amplifies whatever cycle you're in. It made observability, data warehousing, and messaging look like the best businesses in software during the 2020-2022 expansion, and it made the same businesses look structurally impaired during the 2023-2025 efficiency wave. Neither read was correct. If you're designing pricing today, the hybrid shape — a meaningful platform fee plus metered usage above it — buys you a floor in exchange for a lower ceiling. That's a trade many teams should take and few consider before the cycle turns.

Confusing a maturation curve with a broken business. The final pitfall, and the one that produces the worst decisions. Datadog grew roughly $400M of net new revenue in FY25. That is an enormous business performing well. It also decelerated, because those two facts are entirely compatible at scale. The interesting question was never whether the slowdown was real; it was whether Bits AI and Cloud SIEM could become large enough, fast enough, to re-accelerate a $3B base — and whether NRR would hold near 115% or continue sliding toward 110%, which is roughly the line between a durable compounder and a company that has to buy its next growth curve.

Related questions

Is a consumption pricing model still worth adopting after seeing this?

Yes, with eyes open. Consumption aligns price to value and removes seat-count friction, but it transfers cycle risk onto you. A hybrid — platform floor plus metered usage — keeps most of the alignment while putting a floor under revenue in efficiency cycles.

How early can you detect an optimization wave in your own base?

Roughly two to three quarters ahead, if you instrument usage rather than contracts. Watch ingest volume trend, retention-window config changes, and metric cardinality per account. Those move before renewals do, and they move first in your largest accounts.

Did competition actually take Datadog's revenue?

Mostly it applied pricing pressure rather than displacing accounts. Grafana wins on cost in a meaningful share of mid-market evaluations and hyperscaler tools ride existing cloud commitments, but the larger drag was customers optimizing their own usage.

What would signal re-acceleration?

NRR stabilizing or ticking up, $100K+ logo adds returning toward historical pace, and a new SKU crossing roughly $300M ARR. Any one alone is noise. Two together, sustained across three quarters, is a trend.

FAQ

Was the slowdown a sign of a broken business?

No. Growing roughly 24% on a $3B revenue base while adding several hundred million dollars of net new revenue is strong performance at that scale. The deceleration reflected customers optimizing consumption and the arithmetic of a large installed base, not product-market fit erosion or a demand collapse.

How much of the slowdown was competition versus customer optimization?

Customer optimization was the larger driver. Competition from Grafana, a recapitalized New Relic, and bundled hyperscaler tooling showed up mainly as pricing pressure at renewal and as a cost narrative in mid-market evaluations, rather than as wholesale displacement of large accounts.

Why is net revenue retention the metric to watch?

Because on a stable base, growth approximates NRR minus 100%. Falling from above 130% to around 115% mechanically removes roughly fifteen points of growth from retained customers. Everything else — new logos, new SKUs — is working to offset that single line.

Can Bits AI and Cloud SIEM re-accelerate growth?

Eventually, not immediately. Cloud SIEM crossing a $100M ARR run rate is meaningful but small against $3B. New SKUs typically need to reach several hundred million in ARR before they move a company-level growth rate, which is a multi-year timeline rather than a multi-quarter one.

Does this pattern apply to other consumption-priced vendors?

Yes. Any vendor metering volume — data warehousing, messaging, managed databases, log ingestion — carries the same exposure. The same FinOps discipline that pulled hyperscaler growth from above 30% into the low teens propagated through every layer priced on top of that consumption.

What should a RevOps team do differently after watching this play out?

Instrument usage telemetry into the revenue forecast, segment NRR by cohort instead of reporting a blended number, size commit contracts off trailing actuals rather than customer optimism, and separate gross from net retention so you can tell a satisfaction problem from a consumption one.

Sources

• https://investors.datadoghq.com/ — Datadog Investor Relations: quarterly results, shareholder letters, and guidance.
• https://www.sec.gov/edgar/search/#/entityName=Datadog — SEC EDGAR filings including Datadog 10-K and 10-Q reports.
• https://www.datadoghq.com/pricing/ — Datadog's published pricing across infrastructure, APM, logs, and security products.
• https://www.datadoghq.com/blog/ — Datadog engineering and product blog, including DASH conference announcements.
• https://www.gartner.com/en/information-technology — Gartner IT research covering observability and monitoring market dynamics.
• https://www.finops.org/ — FinOps Foundation: frameworks and practices behind the cloud cost optimization wave.
• https://ir.aboutamazon.com/ — Amazon Investor Relations: AWS segment revenue and growth, the upstream consumption signal.
• https://grafana.com/ — Grafana Labs: product and pricing positioning for the open-source-anchored alternative.
• https://newrelic.com/ — New Relic: consumption-based observability pricing and platform positioning.

Related on PULSE

• Why did Snowflake growth slow in 2024-25?
• Why did Salesforce growth slow in 2025?
• Why did HubSpot's customer growth slow in 2025?
• Why did Outreach growth slow in 2024-25?
• How do you run a deal-desk approval workflow that does not slow enterprise deals in 2027?